Investigations of the influence of composition on the properties of refractories are done mainly by the experimental--statistical method. Theoretical approaches to modeling the properties of the goods enable us to obtain only a very wide range of chemical and grain-size composition in the articles.The chemical composition of refractory articles is determined by their classification [I]. However, the search for compositions with the aim of obtaining the optimum properties inside the accepted classification is going on continuously. Alloying is particularly widely used and this considerably improves the properties of the goods.During investigations of the influence of additives on the properties of the articles experimenters avoid the study of a large number of additives. This is because the increase in the number of additives leads to a sharp increase in the number of experiments. However, contemporary methods of planning the experiments enable us to separate the substantially influencing additives using a relatively small number of experiments.The most favorable approach to the study of the influence of additives consists in including, in the first stages, the maximum number of additives, and in separating, with the aid of statistical methods, the substantially influencing additives on the basis of a priori information, or as a result of an experiment.One such method is the random-balance method, the essence of and procedure for which are described in detail in [2,3]. The method was used in the East Institute of Refractories for studying the conditions of obtaining and separating mullite from aluminosilicate materials. The method was used in the first stage of a study for separating addftives substantially influencing mullite formation. The method of proceeding and the results of the experiments are described in [4,5]. As a result of the use of the method of random balance it was possible in 64 experSments, from 26 additives, to separate 7 (CaF2, Na2SiF~, ALF3, Cr203, MgS04, B203, and Ti02) substantially influencing the mullite formation. The response function consisted of the mullite y~eld
M~il(I} m= MfirEd k .lO0M~uit (2) where Mmull is the mass of mullite formed, g7 Mfired k~ mass of fired kaolinite, g; and MT mull, theoretically possible yield of muilite, g.After identifying the substantially influencing additives for further investigation, other methods of planning the experiments are used. In particular, widely used are the methods of complete and fractional factor experiments (PFE and DFE) [2,3]. These methods were also used for further investigation of the influence of additives on the mullitization of kaolin.To construct the plan of the experiment we used 1/4 of the PFE replica type 27-2 . The planning matrix, the results of the experiment, and the regression coefficients are given in [4]. As a result, after 32 experiments it was possible to obtain an adequate regression equation and make conc%usions about the influence of the mentioned additives. It turned out that the additives AiF3, Cr203, and TiO2' m...